Abstraction Refinement via Inductive Learning

Alexey Loginov, Thomas Reps, and Mooly Sagiv

This paper concerns how to automatically create abstractions for program analysis. We show that inductive learning, the goal of which is to identify general rules from a set of observed instances, provides new leverage on the problem. An advantage of an approach based on inductive learning is that it does not require the use of a theorem prover.

(Click here to access the paper: PostScript, PDF, (c) Springer-Verlag)